Performance of a Probabilistic Cloud-to-Ground Lightning Prediction Algorithm

A probabilistic cloud-to-ground lightning algorithm was created by

training a neural network on storm characteristics. The input dataset

consisted of all storm cells over the entire coterminous United States

on 12 days in 2008-2009 (one day per month). The input characteristics

include radar and near-storm environmental parameters and the neural

network was set up so that its output is the probability of cloud-to-ground

lightning at a grid location 30 minutes in the future. The probabilistic

output was evaluated on several independent test dates in 2009 and

results of that evaluation are presented.